How to use analytics in business - the right way

Last week I had a chance to attend the Predictive Analytics Innovation Summit in San Diego, as well as give a talk to the audience about how LinkedIn uses analytics. As a passionate follower of the subject matter it was great to share ideas with other industry professionals as well as learn about new trends in analytics aimed to solve the most complicated and subjective problems. For example, analytical techniques are being developed to construct an objective answer to a subjective question "who is the best basketball player of all time?"

While analytics seem like a solve for everything (especially in marketing, where everything can now be quantified vs. previously had to live with lots of 'fluffy stuff'), it was eye opening that analytics, if not applied correctly, has a danger of leading you to a wrong decision. This is quite alarming because making a 'data-driven decision' is supposed to be extremely objective, and should lead you in making the most informed and appropriate action. Let's look at a few examples:

• At Company X, they found that extroverts are more successful => They should stop hiring introverts
• At Company Y, they found that people acquired through promotions exhibit higher churn => They should stop promotions
• At Company Z, they found that rated-R movies have higher regression coefficient than PG-13 movies in retail sell-through => They should make only rated-R movies

With such extreme setting the above scenarios may sound silly, but as these examples get more nuanced the analytical output may not be sufficient in guiding you to an optimal action. Synthesizing the learnings and insights from the conference regarding this issue, I've found that having a few principles in mind can help you avoid falling into this unfortunate situation.

1. Solve the right problem

Analytical methods bring scientific and objective framework in answering a question. However if the question is wrong, the answer does not matter. Company Y's ultimate business question is not to lower churn rate. It is to maximize business results by managing business drivers such as churn. It is extremely important to be absolutely clear on the problem statement so we can have the right 'target variable' to model against.

2. Pick a right approach

Once a problem is clearly defined, we need to pick a right approach to solve the problem. Unlike pure mathematics, real-life business problems often has no one solution set but many, and likewise multiple ways to get to an answer. While many of us may think A/B testing is the golden analytical approach, sometimes that is not the case. For example, social networking sites have extremely high virality and network effects that make A/B testing difficult. Facebook's Look Back campaign has been one of their most successful marketing initiatives in driving incremental engagement. Due to its viral nature, there was no way to have a control group where they didn't get the campaign vs. those who received. So the folks in Facebook took a different approach for measuring incrementality driven from this initiative. Should they have stuck to a traditional A/B testing, not only would analysts have been so devastated but it also could have resulted in a sub-optimal user experience. How would you have felt when you couldn't get the cool feature that your friend got?

3. Speak the same language

Lastly, it is so important that business users are grounded with an analytical mindset, as well as analytics professionals ramp up on business acumen. So many times the business users are making wrong interpretations of the analytical output (see above examples), as well as analysts being unable to provide the business with the right answer, or fail to communicate the right answer (i.e., the 'so-what'). Make sure to invest time up-front in having the two cultures gel and understand how the business and analytics can be in-tune. Business and analytics 'oscillating at the same frequency' will set the team up nicely for answering the most complex business problems with the most elegant analytical rigor.

I'm sure there are more things to consider, but I hope these principles can help the business embrace analytics in a right way to make the better decisions.

* PS - below is what I presented at the conference with my colleague, Wenjing Zhang.

Photo: Wenjing and myself presenting at the Predictive Analytics Innovation Summit in San Diego, Feb. 13, 2015. (credit: Xin Fu, Director of Data Science at LinkedIn)